top of page

Performance Management Agent

Reframing performance management from administrative detective work to continuous, AI-assisted coaching.

Agentic AI · Workflow Strategy · Enterprise UX

I explored how an AI agent could reduce the administrative burden on managers by gathering context, surfacing what matters, and guiding the next action while preserving review and human control.
AT A GLANCE
ROLE

Senior UX Design Engineer

TOOLS

Figma · React · TypeScript · Cursor · GitHub · Workday design-system components

TIMELINE

August 2025 - July 2026

PARTNERS

Product · Engineering · Design · Research · Leadership

SCOPE

Performance Management · Manager workflows · Goals · Feedback · Reviews · Coaching

WHAT I OWNED

Agent interaction model · Workflow simplification · Context and recommendations · Human review · High-fidelity prototyping

CHALLENGE

Performance management can become a high-stakes administrative exercise in which managers spend substantial time gathering information, reconstructing prior conversations, checking goals, finding feedback, and preparing for reviews before they can actually coach someone.


The design problem was not simply to make forms faster. It was to reduce the amount of detective work required before a manager could make a useful decision or have a meaningful conversation.

STRATEGIC SHIFT

The opportunity was to move from software that stores performance information to software that actively helps make sense of it.


The system should do the gathering; the manager should keep the judgment.


An agent could synthesize context, identify gaps, suggest next actions, and prepare the manager for a conversation. But the product still needed to make its reasoning visible and keep consequential decisions reviewable.

MY ROLE

Automate preparation before judgment

Some of the most valuable AI opportunities sit before the decision, not inside it.


Trust requires visible reasoning

If a system influences a high-stakes conversation, users need enough context to understand why the recommendation exists.


Agentic UX is about responsibility

The core design question is not simply what an agent can do, but what it should do—and who remains accountable afterward.

KEY DECISIONS

Decision 1 — What work should the agent remove?

The highest-value opportunity was not automating the final judgment. It was removing the repeated preparation work around that judgment: gathering context, surfacing changes, and identifying what required attention.


Decision 2 — What should be a recommendation versus an action?

The more consequential the decision, the more important it became to preserve review and control. I treated AI as a system for preparation and guidance first, rather than silent autonomous execution.


Decision 3 — How do you make the system trustworthy?

Recommendations needed enough context for the manager to understand why something was surfaced. Trust depended on traceability, not just confidence in the model.

HOW I WORKED

The concept shifted the focus from completing periodic performance-management tasks toward reducing the cognitive and administrative burden surrounding them.


It established a clearer model for agentic assistance: gather the context, surface what matters, recommend what to do next, and keep the human responsible for consequential decisions.

WHAT CHANGED

The concept shifted the focus from completing periodic performance-management tasks toward reducing the cognitive and administrative burden surrounding them.


It established a clearer model for agentic assistance: gather the context, surface what matters, recommend what to do next, and keep the human responsible for consequential decisions.

LEARNINGS

Automate preparation before judgment

Some of the most valuable AI opportunities sit before the decision, not inside it.


Trust requires visible reasoning

If a system influences a high-stakes conversation, users need enough context to understand why the recommendation exists.


Agentic UX is about responsibility

The core design question is not simply what an agent can do, but what it should do—and who remains accountable afterward.

CONTRIBUTIONS

Product Design

Performance workflows · Manager experience · AI-assisted guidance · Coaching


Agentic UX

Recommendations · Context synthesis · Human review · System behavior


Design Engineering

React prototyping · TypeScript · AI-assisted development · Working interaction models


Systems Thinking

Goals · Feedback · Reviews · Development · Manager decision support

port·fo·li·o| pôrtˈfōlēō | noun (plural portfolios) - a large, thin, flat case

​

© 2026 by Michael L. Olivier | thinflatcase.com

  • Medium
  • GitHub
  • LinkedIn
bottom of page